An exact rule raised causal 6 Hz SEEG forecast acceptance by 2.23 percentage points in 21 patients under the stated 20 ms RMS error model

An exact angle rule accepted more outputs from a causal 6 Hz SEEG phase reader than a matched disk rule. We used public OpenNeuro ds006519 version 1.0.0. The test had 35 records, 21 patients, and 1,754 source-good signal occurrences. The reader used a 500 ms past window and an assumed 50 ms delay. The primary error model allowed real voltage errors only in the final 20 ms. Their RMS was at most one four-second signal standard deviation. The scale stayed fixed at its nominal value. The phase change limit was 45 degrees. Equal-patient mean acceptance was 87.38% for the exact rule and 85.15% for the disk rule. The gain was 2.23 percentage points. The descriptive 95% patient bootstrap interval was 1.96 to 2.49 points. All 21 patients had a positive gain. The proof gives the exact strict angle condition for a fixed linear reader and a real Euclidean error ball. Direct projections matched all 898,048 primary decisions. Fresh source extraction matched all 595 preserved arrays. The guard limits output change under the stated error model. It does not establish accurate brain-phase prediction. Exact accepted forecasts differed from an offline 4–8 Hz reference by 81.32 degrees on average. Full raw erasure exceeded the assumed error limit in 49.21% of points. The whole-epoch scale is retrospective. No device or delivered stimulation was tested. The package preserves code, proof, compact inputs, native voltage examples, source hashes, results, and three figures. The full raw data have a checked retrieval recipe. Related exact phase geometry is credited; we claim a new benchmark, not a new general theorem.

signal processing · phase estimation · brain-computer interface · SEEG · robustness certificate · error bound · causal filter · intracranial EEG · phase accuracy · OpenNeuro